Merger Effects with Product Complementarity: Evidence from
Colombia’s Telecommunications

Juan Vélez

February 04, 2017

Intro

Thanks, it is an honor to be here
First time presenting to Colombian audience
Work in progress, comments welcome
I'll talk about mergers with potential complements, let me start by defining…

Product complementarity

Two ways one can think of complementarity

Cross-price elasticity is negative
Joint consumption yields extra utility
Text book example: coffee and sugar.
- A decrease in the price of coffee increases demand for sugar

The two definitions are really the same.
I don't mean complements in the production sense
- Actually I abstract away from it

Mergers with product complementarity

Complements: consumers get extra utility from joint consumption
- standalone consumption is possible
The complementarity creates pro-competitive effects
Bundling could create anti-competitive effects
Example: coffee and sugar
- Reduction in the price of coffee increases profits for sugar producer
- The merger internalizes those extra profits (pro-competitive)
- Merged firm can bundle and use it for price discrimination (anti-competitive)
Different than horizontal and vertical mergers

Twofold interest: dual character, regulatory agencies
Horizontal: produce substitutes (Staples and Office Depot, Bavaria and Águila)
Vertical: components of a final good (Time Warner and Turner Corp, Gabrica and Zoetis)

Recent examples: smartphones and headsets

2014, 3.2 billion dollars (9.6e12)
Beats: 2 business (music and electronics)
Authorities worried about music and not electronics

Recent examples: phone and TV

2015 67 billion (1.8e13)
AT&T second largest telecom

Recent examples: Internet and cable TV

AT&T strikes back in 2016
85 billion (Time Warner is cable TV)
It hasn't happened yet
Trump opposes to it because he i afraid of media and it consolidates media power

Recent examples: Comcel and Telmex

Sleaziest merger in history
involved larges carrier and second largest ISP
Comcel market shares have dwindle down since 2009

Recent examples: ETB and …

Still don't know if it will be a merger of complements
Some complementarity is bound to happen
These kinds of mergers happen often and affect lots of consumers

Motivation

Vast literature studying mergers
- Vertical
- Horizontal
Regulators around the world have handbooks that describe the scrutiny process.
- Horizontal Merger Guidelines (2010) in the US
- Ley 1340 de 2009 in Colombia
Mergers with complementary goods happen often
- Telmex-Comcel, Tigo-Une, Telecom-Movistar…
Technological convergence (TV used to be independent from Internet)

Surge in mergers with complements
Technological convergence
guidelines define how to scrutinize and remedies to problems caused by allowed mergers
- Divestiture
- Forbid bundling
Bundling might be actually a good thing if the complementarity is strong

Overview

I study mergers of firms producing (potentially) complementary products.
- Coffee and sugar, smartphones and earbuds, cellphone and Internet.
Why should we care?
- Theory predictions are ambiguous (Choi, 2008; Anderson et al., 2010)
- Often happen: Apple-Beats, AT&T-DirecTV, AT&T-Time Warner
This paper
- Estimates a discrete choice demand (substitution patterns)
- Finds that wireless and wired services are complements (caveats)
- Uses preference estimates to evaluate 2 alternative scenarios
  - Merger of ETB and Avantel
  - De-merger of Claro into Telmex and Comcel

I don't know if goods are complements a priori (estimate demand, get )

Relation to literature

Theory papers
- Spector (2007)
- Peitz (2008)
- Choi (2008)
- Anderson, Loertscher, and Schneider (2010)
- Flores-Fillol and Moner-Colonques (2011)
Empirical papers
- Gentzkow (2007)
- Ribeiro and Vareda (2010)

Spector shows bundling can be used to sustain collusive behavior
Choi study effects of mergers on price, investment and R&D
Peitz circumstances under which bundling can blockade entry
Anderson builds on Choi: when can a merger benefit consumers, harm outsiders and benefit insiders
All assume no cost energy

Data and industry

Colombia's telecom sector: the timeline

Crash course on history of telecom
I changed my perception about the sector after reading more
Developed as they normally do.
91 increased competition
Claro has been losing market shares
VNMOs have increased competition

Colombia's telecom sector: the players

Characterized by a few big firms (less than 10)
- Telephone & Internet: Claro, Movistar, UNE, ETB and Emcali.
- Mobile: Claro, Movistar, Tigo, Avantel.
- TV: Claro, DirecTV, ETB, Movistar, UNE.

A fringe of small ones (more than 70).

The actual market structure is also similar to those in developed nations. Certainly better than the region. (16 billion revenues)
Small ones are exclusively wired
VNMO are changing the landscape

Data

I use data from two sources

Data provided by the Comisión de Regulación de Comunicaciones (CRC)
- Firms fill out Form 5 at the end of every quarter.
- Market shares and bundle characteristics.
- 762 markets (3,048 market-quarters).
- 82 firms.

I use 2 datasets
Observe universe of firms and products
Amazing amount of detail on good's characteristics
I do drop some small towns

Data: Form 5

Decree defines information to be collected
Astounding amount of detail
I observe standalone and bundles
Location up to stratum, price, and many characteristics
All these characteristics will help identify complementarities later

Data Household survey

Data on households (CPS-like, quarterly)
- mean schooling
- household head between 25-45 years old
- monthly income
- family size
(Households summary statistics)

The other data I use is GEIH
Sample households corresponding to each market
This data will allow me to let households with different observed characteristics to respond differently to bundle's attributes
Rich less price elastic, educated diff attitude toward premium channels, Older households may not care about download speed.

Summary statistics: firms

	2015 - Q1	2015 - Q2	2015 - Q3	2015 - Q4
Summary statistics of firm attributes Means and standard deviations
Firm level
# plans ^a	22.831	24.12	21.965	22.631
	21.124	22.167	19.765	18.149
# cities ^b	24.312	24.111	27.304	27.125
	20.651	21.31	22.651	24.435
Observations	82	82	81	79
Firm-market level
Price ^c	19.203	19.605	21.317	22.481
	14.319	16.129	15.874	15.104
# plans ^d	6.819	6.617	6.429	6.597
	6.776	6.443	5.601	6.335
# subscribers	32341.879	32346.814	32410.612	32496.101
	31516.417	30478.309	32271.051	33064.021
Observations	2341	2346	2347	2351
_{a: total plans offered across markets. b: number of cities where firm operates. c: price of bundle in 2015 dollars. The number of households subscribed is used as weights. d: number of plans offered in market.}

Main takeaway is that the firms are VERY diverse (table isn't much useful)
Standard deviations are huge! because of the wide range of many of these variables
Average firm 20 plans, with some ISPs having very few (Eje Cafetero 2 plans)
Define what a plan is: change in characteristics define a different plan
Average firm operates in 24 cities, but mobile operate almost everywhere and lots of small ISPs

Summary statistics: plans

	2015 - Q1	2015 - Q2	2015 - Q3	2015 - Q4
Summary statistics of plan attributes Means and standard deviations
Price ^a	26.388	29.718	28.31	27.55
	21.127	28.271	21.877	22.101
Download speed ^b	2.285	3.121	3.426	3.375
	1.921	2.815	3.107	3.264
Premium channels ^c	0.312	0.336	0.345	0.388
	0.304	0.325	0.318	0.342
HD channels	0.265	0.211	0.293	0.238
	0.187	0.177	0.265	0.148
Gb Included (Mobile) ^d	3.421	3.678	3.762	3.518
	2.901	2.875	2.886	3.107
Minutes included (Cell phone)	301.078	409.123	410.356	452.31
	100.078	102.891	132.376	137.552
Gb included (Cell phone) ^e	2.678	2.889	3.173	3.204
	4.143	5.168	5.284	4.178
Minutes included (Phone)	435.218	482.337	473.215	461.461
	212.166	283.472	251.639	258.813
Observations	28934	28965	28919	28727
_{a: Price in 2015 dollars. b: of bundles with Internet. In megabits per second. c: channels like HBO. For bundles with TV. d: mobile refers to mobile internet provided through a dongle. e: access to internet using a cell phone as opposed to using a USB modem and a computer.}

Average plan is around 26 dollars
Download speeds have been increasing but don't match those reported by the Ministerio de Telecomunicaciones
A third of cable TV include some premium channels

Summary statistics: correlations

Remember sugar and detergent?

For my purpose this one is more interesting
Blue 'happen together, red 'either is bought'
Symmetric of course. Just focus on the top left half
suggest that Mobile internet and Cellphone are complements: maybe just available together
mobile and phone substitutes? not necessarily (mobile may be used just coz broadband not available)
Good first approach though

Demand

Estimating substitution patterns

Following Gentzkow (2007)

Suppose two goods A and B.

\begin{matrix} (1) & u_{A} = δ_{A} - α p_{A} + ξ_{A} \end{matrix}

$u_A=\delta_A - \alpha p_A + \xi_A$

\begin{matrix} (2) & u_{B} = δ_{B} - α p_{B} + ξ_{B} \end{matrix}

$u_B=\delta_B - \alpha p_B + \xi_B$

\begin{matrix} (3) & u_{A B} = u_{A} + u_{B} + Γ \end{matrix}

$u_{AB}=u_A + u_B + \Gamma$

$\delta$ : mean utility
$\xi$ : unobservable variation in utility.
$\Gamma$ : utility/dis-utility of joint consumption
Sign of $\Gamma$ determines if complements or substitutes (more details).

Gentzkow motivation: are online newspapers substitutes for online?
bundles are modeled as just another good available to consumers
2 goods, 4 possible bundles: A, B, AB, outside option
Utility 3 components: good specific unobserved, price and mean utility
Utility of bundles: sum of utilities + or - something
Sign of Gamma determines subst or complts

Empirical demand I

ℱ firms indexed by $f\in\{1,2,3,...,ℱ\}$
$g \in \{1,2,3,...,G_f\}$ indexes the standalone goods provided by a firm
$b \in \{1,2,3,...,ℬ(G_{f})\}$ indexes bundles provided by a firm
$j \in \{0,1,2,3,...,2^{\max\{G_1,G_2,...,G_ℱ\}}\}$ are types of bundles
In the application there are 32 possible bundles
- outside bundle is 0
- singleton bundles are 1-5
Example: Sutatausa. ℱ = 4 (Claro, UNE, Coltel and ETB). Claro provides TV, Mobile Internet and Cellphone. Bundles: TV/Mobile (1.2Gb), TV/Mobile (2Gb), Cellphone/TV).

I need to adapt to more than 2 goods

Empirical demand II

Utility of a single good $g$

\begin{matrix} (4) & {\bar{u}}_{i g} = p_{g} {\bar{α}}_{i} + \sum_{k} x_{g k} {\bar{β}}_{i g k} + ξ_{g} \end{matrix}

$\bar{u}_{ig} = p_{g} \bar{\alpha}_i + \sum_k x_{gk} \bar{\beta}_{igk} + \xi_{g}$

\begin{matrix} (5) & {\bar{α}}_{i} = α + \sum_{r} z_{i r} α_{r}^{o} + ν_{i p} α^{u} \end{matrix}

$\bar{\alpha}_i = \alpha + \sum_r z_{ir}\alpha_r^o + \nu_{ip} \alpha^u$

\begin{matrix} (6) & {\bar{β}}_{i g k} = β_{g k} + \sum_{r} z_{i r} β_{g k r}^{o} + ν_{i g k} β_{g k}^{u} \end{matrix}

$\bar{\beta}_{igk} = \beta_{gk} + \sum_r z_{ir}\beta_{gkr}^o + \nu_{igk} \beta_{gk}^u$

$k$ are non-price characteristics of the good.
$r$ are characteristics of an individual (household).

In principle one can run a logit (alpha u and alhp o == 0)
Equivalent to say that poor and rich households respond to price
Infamous for messing up substitution patterns

Empirical demand III

utility of a bundle $b$ of type $j$

\begin{matrix} (7) & u_{i b} = {\begin{cases} ε_{i b}, & if b = 0 \\ \sum_{g \in b} {\bar{u}}_{i g} + Γ_{j (b)} + ε_{i b}, & if b > 0 \end{cases} \end{matrix}

$u_{ib} = \begin{cases} \varepsilon_{ib}, & \text{if } b=0 \\[2ex] \sum_{g \in b} \bar{u}_{ig} + \Gamma_{j(b)} +\varepsilon_{ib}, & \text{if } b >0 \end{cases}$

$\varepsilon_{ib} \sim$ T1EV
$\Gamma_{j(b)} = 0$ for $b < 5$

Now we can define the bundle's utility (bundle includes outside! and singleton)
So the shocks are realized and consumers choose which bundle
Explain why T1EV
Get an expression of how likely it is that a consumer will choose j (market share)

Empirical demand IV

Integrating over $\varepsilon$

\begin{matrix} (8) & s_{b} = \int_{ν} \int_{z} \frac{\exp [\sum_{g \in b} {\bar{u}}_{i g f} + Γ_{j (b)}]}{1 + \sum_{f = 1}^{ℱ} \sum_{l = 1}^{ℬ (G_{f})} [\exp \sum_{g \in l} {\bar{u}}_{i g} + Γ_{l (b)}]} d F_{z} d F_{ν} \end{matrix}

$s_{b} =\int_\nu \int_z \frac{\exp \left[\sum_{g \in b} \bar{u}_{igf}+ \Gamma_{j(b)} \right]}{1 + \sum_{f=1}^{\text{ℱ}} \sum_{l=1}^{\text{ℬ}(G_f)} \left[\exp \sum_{g\in l} \bar{u}_{ig} + \Gamma_{l(b)} \right] }dF_z dF_\nu$

Follow Berry (1994)
Hausmann instruments and BLP instruments
GMM
- $E [Z_m\omega(\theta)]=0$

integrating epsilon out, we get an expression of market shares in terms of parameters of interest in particular GAMMA
Basic idea? find Gamma that sets estimated shares equal to osberved
Why not do it? Parameters enter non linearly. With quantum computers we could.
Berry found a way (3steps): 1 pick arbitrary Theta2, do the berry iteration, define omega = delta- linear construct GMM objective

Identification of Γ

Suppose $s_A$ , $s_B$ and $s_{AB}$ are observed

Large relative to and could indicate they are complements ()
- but also that they are bad substitutes (sugar and detergent)
- or correlated preferences (cinephiles)
To get true complementarity I need exclusion restrictions
- Suppose variable $x$ enters $\delta_A$ but not $\delta_B$ nor $\Gamma$
- observing $x'$ adds three pieces of information: $s_A(x')$ , $s_B(x')$ and $s_{AB}(x')$

Berry has proofs showing parameters are formally identified.
I want to add the intuition of why they are identified
Large market s_AB are not enough
- Terrible substitutes (sugar and detergent)
- Correlated preferences(cinephiles)
Key identification: exclusion restrictions (variables included in the mean utility of one of the goods but not on the other's)
Observing 2 values of such variables formally identifies: adds 3 moments and just one additional parameter.

Identification of Γ: intuition

Suppose

That the goods are Internet (A) and TV (B)
Large $s_{AB}$ relative to $s_A$ and $s_B$
The exclusion restriction is download speed (DS)
- If $\Gamma>0$ , $\uparrow DS$ should increase $s_A$ and $s_B$
- If $\Gamma=0$ , $\uparrow DS$ should not alter $s_B$

Internet (A) and TV (B) and true Gamma >0. increase in download speeds should also affect S_B
If gamma = 0 shouldn't
Or think of 2 markets where speed varies. If the market with higher speed has also larger shares for TV and bundle, the model has to rationalize that with a large Gamma

Endogeneity and instruments

Unobserved characteristics may be correlated with price

Has premium channels ☑ HBO? Sundance? Both? Are people renting modems?
In any case, more expensive bundles are (unobservably) more attractive

Instruments

sum of characteristics of other products produced by the firm (BLP)
price of similar products by the same firm in other markets (Hausman)

Usual concern? What identifies the price-slope? demand shifts? supply shifts?
Usually get positive slope demands when unobersved characteristics are correlated with price
Plans with more premium channels tend to be more attractive and more expensive

Instruments: examples

Valid: correlated with prices
Exogenous: uncorrelated with unovserbable utility

BLP instruments

Price of ETB's Plan Trío in Bogotá is instrumented with characteristics of Triple Play plans offered by Claro, UNE and Movistar.

Hausman instruments

Price of UNE's Silver 5 Megas in Medellín is instrumented with prices of the same plan in Rionegro, Bello, Envigado and Sabaneta.

good: correlated with price
uncorrelated with unobserved utility
BLP exo comes from timing
Hausman: after controlling for demographic market shocks are uncorrelated

Supply side

Firms are assumed to play a static Bertrand game

Differentiated goods: characteristics are decided before the game starts
Each firm decides a vector of prices after observing its rivals' poducts' characteristics
Under such conditions, the firms' FOC for a market are given by

\begin{matrix} (9) & p = m c + Ω (p)^{- 1} s (p) \end{matrix}

$\mathbf{p} = \mathbf{mc} + \Omega(\mathbf{p})^{-1}\mathbf{s}(\mathbf{p})$

- $\mathbf{O}$ is $\mathscr{I}\times \mathscr{I}$ with $(m,n)$ element 1 if products $m$ and $n$ are sold by the same firm.
- $\nabla_\mathscr{s}(\mathbf{p})$ the price derivatives of the shares
The Bertrand-Nash equilibrium is the $\mathbf{p}$ that solves the system

Why assume Bertrand? Known FOC
Size of P is numbre of goods in market
Matrix Omega prodcut of 2

How to do counterfactuals

To alter market structures just change $\mathbf{O}$ accordingly and solve for the new equilibrium

\begin{matrix} (10) & p^{p o s t} = \hat{m c} + {[O^{p o s t} \nabla s (p^{p o s t})]}^{- 1} s (p^{p o s t}) \end{matrix}

$\mathbf{p^{\mathbf{post}}}=\mathbf{\widehat{\mathbf{mc}}}+\mathbf{\left[\mathbf{\mathbf{O}^{\mathbf{post}}\mathbf{\nabla s(\mathbf{p^{post}})}}\right]^{-1}}\mathbf{s}(\mathbf{p^{post}})$

Example:

\begin{matrix} (11) & O = [\begin{matrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{matrix}] \to O^{p o s t} = [\begin{matrix} 1 & 1 & 0 \\ 1 & 1 & 0 \\ 0 & 0 & 1 \end{matrix}] \end{matrix}

$\mathbf{O} = \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix} \rightarrow \mathbf{O}^{\mathbf{post}} = \begin{bmatrix} 1 & 1 & 0 \\ 1 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix}$

Computanionally involved but now we can reap the rewards
Estimation is hard, but simmulation is easy
Example

Results

Results: product characteristics

	OLS		BLP Instruments		Hausman Instruments		Random Coef.
Estimates Bundle characteristics
Price	-0.163	**	-0.381	**	-0.411	**	-0.479	**
	0.081		0.154		0.178		0.224
Download speed	2.235	**	1.571	**	1.792	**	2.012	*
	1.047		0.648		0.841		1.088
Premium channels	-0.065		0.031		0.028		0.091	*
	0.067		0.045		0.029		0.046
HD channels	0.078	*	1.318	**	1.285	**	1.247	*
	0.069		0.661		0.517		0.661
Gb included (Mobile)	1.046	**	1.048	**	1.035	**	1.037	**
	0.482		0.511		0.456		0.496
Minutes included (Cell)	0.045	***	0.029	**	0.035	**	0.027	**
	0.012		0.013		0.012		0.012
Gb included (Cell)	0.468	***	0.402	**	0.395	**	0.673	**
	0.114		0.2		0.197		0.285
Minutes included (Landline)	-0.035		0.018		0.018		0.017
	0.231		0.086		0.077		0.45
Constant	-12.489	***	-12.952	***	-13.457	***	-14.357	***
	0.315		1.47		1.057		2.714
_{Continuous variables in logs. Number of osbervations: 115,545 Signif. codes: 0.001(*) 0.05() 0.10(*).}

around 10% inelastic demands
Both isntruments reveal the existence of a downard bias
Instruments solve the problem of this weird negative sign on minutes included
logit gets the wrong substi patterns

Results: Susbtitution patterns

	Γ		Dollars
Estimated substitution patterns
Internet/Phone	0.82	***	$0.92
	0.13
Internet/TV	-1.78	*	-$2.00
	1.6
Internet/Mobile	1.35	***	$1.51
	0.27
Internet/Cell	2.79	**	$3.13
	1.38
Phone/TV	0.92	***	$1.03
	0.04
Phone/Mobile	-2.11	***	-$2.36
	0.57
Phone/Cell	1.02	***	$1.15
	0.02
TV/Mobile	0.02		$0.02
	0.01
TV/Cell	1.73	***	$1.94
	0.01
Mobile/Cell	2.57	*	$2.89
	1.22
Internet/Phone/TV	1.17	***	$1.31
	0.26
_{The estimation algorithm is initialized with simple Logit estimates. * significant at 1%; significant at 5%}

All signigifacnt except for TV-Mobile

Results: Susbtitution patterns

	Γ		Dollars
Estimated substitution patterns
Internet/Phone	0.82	***	$0.92
	0.13
Internet/TV	-1.78	*	-$2.00
	1.6
Internet/Mobile	1.35	***	$1.51
	0.27
Internet/Cell	2.79	**	$3.13
	1.38
Phone/TV	0.92	***	$1.03
	0.04
Phone/Mobile	-2.11	***	-$2.36
	0.57
Phone/Cell	1.02	***	$1.15
	0.02
TV/Mobile	0.02		$0.02
	0.01
TV/Cell	1.73	***	$1.94
	0.01
Mobile/Cell	2.57	*	$2.89
	1.22
Internet/Phone/TV	1.17	***	$1.31
	0.26
_{The estimation algorithm is initialized with simple Logit estimates. * significant at 1%; significant at 5%}

pairs higlihted show extra utility of joint consumption and their valuation
Beyond the point estimates the takeway is that, in general, wired and wireless are complements.

Counterfactuals

Alternative scenarios

With estimates for the preferences, we are ready to simulate 2 scenarios:

Merger of ETB (Wired) and Avantel (Wireless)
- Allows to see the effect on price of standalone goods.
- Doesn't capture bundling.
De-merger of Claro into Telmex (Wired) and Comcel (Wireless)
- Captures the effect of bundling and price discrimination.
- Helps scrutinize merger in hindsight.

Interest for policy makers
avantel is the last exclusive mobile carrier
claro: previous simulation doesn't capature effect of bundling

Merger ETB and Avantel

Slight increase price of standalone
reduction price of bundles(NOT from the merged!!!!)

Merger ETB and Avantel (cont.)

	1st Quartile	2nd Quartile	3rd Quartile	Mean	Min.	Max.
Prices of standalone goods
Baseline^a	$7.79	$10.40	$16.01	$13.68	$1.03	$49.28
Merger^b	$8.23	$10.72	$16.31	$13.72	$0.95	$49.21
_{(a) Original market structure; (b) ETB and Avantel.}

	1st Quartile	2nd Quartile	3rd Quartile	Mean	Min.	Max.
Price of bundled goods
Baseline^a	$14.16	$17.02	$25.88	$20.28	$8.31	$153.83
Merger^b	$13.33	$14.59	$22.55	$18.49	$7.15	$153.97
_{(a) Original market structure; (b) ETB and Avantel.}

The average households gets a little over one extra dollar worth of consumers surplus under the new price distribution.
The total gain in consumer surplus for the 13 largest cities is over 7 million.

1st and 3rd quartile increase as well as mean (min and max cheaper)
More interesting than the price distribution is the effect on consumer surplus: 7 million dollars for 13 largest MSA

De-merger of Claro

Largest carrier and second largest ISP
Sneakiest move in the history of mergers

De-merger of Claro (cont.)

	1st Quartile	2nd Quartile	3rd Quartile	Mean	Min.	Max.
Price of standalone goods
Baseline^a	$7.23	$9.72	$15.30	$12.90	$0.05	$49.12
Merger^b	$6.79	$9.40	$15.01	$12.86	$0.03	$49.13
_{(a) Original market structure; (b) ETB and Avantel.}

	1st Quartile	2nd Quartile	3rd Quartile	Mean	Min.	Max.
Price of bundled goods
Baseline^a	$18.48	$28.60	$37.62	$28.57	$5.99	$158.18
Merger^b	$20.72	$28.96	$37.45	$29.42	$6.15	$158.22
_{(a) Original market structure; (b) ETB and Avantel.}

The consumer surplus is reduced by a little more than $11 million.

As expected there is a slight reduction in standalone prices but the price of bundles increases
Consumer variation is a negative 11 million

Final remarks

Extensions

Allow for cost synergies.
Endognenous product characteristics as in Draganska et al. (2009) and Fan (2013).
Switching costs (Lock-in)

Policy implications

Stronger complementarities require less stringent scrutiny
Allow bundling
Investment on complements as alternative to divestiture (horizontal mergers)

Future work

Rule of thumb for regulators
Endogenous product characteristics
Price discrimination
VNMO

Thanks!

Summary statistics: households

	Stratum 1	Stratum 2	Stratum 3	Stratum 4	Stratum 5	Stratum 6
Summary statistics of household characteristics Means and standard deviations
Avg. Schooling ^a	6.94	8.28	9.72	11.75	11.85	13.86
	2.45	2.49	2.96	3.42	3.27	3.42
HH age ^b	0.68	0.61	0.62	0.59	0.56	0.65
	0.47	0.49	0.49	0.49	0.5	0.48
Family size	5.34	4.76	3.65	3.86	3.59	3.37
	3.45	2.27	1.48	1.62	1.53	1.52
Monthly income ^c	386.26	534.17	556.37	933.36	1084.72	1957.3
	366.42	363.72	523.72	997.25	1549.46	2277.01
Observations	720	720	720	720	390	390
_{a: sum of the number of years of schooling for people in the household, devided by the number of members of the household. b: head of the household is between 25 and 45 years old. c: Monthly income in 2015 dollars}

(go back)

Choice probabilities

Let $\mathbf{u}=(u_A,u_B,u_{AB}) \sim F(\mathbf{u})$ . Then, the choice probabilities are:

\begin{matrix} (12) & q_{A} = \int_{u} I (u_{A} \geq 0) I (u_{A} \geq u_{B}) I (u_{A} \geq u_{A B}) d F (u) \end{matrix}

$q_A = \int_u I(u_A \ge 0)I(u_A \ge u_B)I(u_A \ge u_{AB})dF(\mathbf{u})$

\begin{matrix} (13) & q_{B} = \int_{u} I (u_{B} \geq 0) I (u_{B} \geq u_{A}) I (u_{B} \geq u_{A B}) d F (u) \end{matrix}

$q_B =\int_u I(u_B \ge 0)I(u_B\ge u_A)I(u_B \ge u_{AB})dF(\mathbf{u})$

\begin{matrix} (14) & q_{A B} = \int_{u} I (u_{A B} \geq 0) I (u_{A B} \geq u_{A}) I (u_{A B} \geq u_{B}) d F (u) \end{matrix}

$q_{AB} =\int_u I(u_{AB} \ge 0)I(u_{AB}\ge u_A)I(u_{AB} \ge u_B)dF(\mathbf{u})$

Let $Q_A = q_A+q_{AB}$ and $Q_B = q_B + q_{AB}$ .

Definition: A and B are complements if $\frac{\partial Q_A}{\partial p_B} <0$ , susbtitutes if $\frac{\partial Q_A}{\partial p_B} >0$ and independent if $\frac{\partial Q_A}{\partial p_B} = 0$ .

Substitution patterns: $\uparrow p_B$ and $\Gamma = 0$

marginal consumers like $m$ will substitute $AB$ for $A$ ,
$\uparrow q_A = \downarrow q_{AB}$ and $Q_A$ remains unchanged.
marginal consumers like $n$ have no impact on $Q_A$
Hence $\Gamma = 0$ implies $\frac{\partial Q_A}{\partial p_B}=0$ (goods are independent)

Substitution patterns: $\uparrow p_B$ and $\Gamma > 0$

marginal consumers like $m$ will substitute $AB$ for $A$
$\uparrow q_A = \downarrow q_{AB}$
marginal consumers like $o$ will substitute $AB$ for neither good
$\downarrow q_A$ and overall $\downarrow Q_A$
Hence $\Gamma > 0$ implies $\frac{\partial Q_A}{\partial p_B}<0$ (goods are complements)

Substitution patterns: $\uparrow p_B$ and $\Gamma < 0$

marginal consumers like $m$ will substitute $AB$ for $A$ ,
$\uparrow q_A = \downarrow q_{AB}$
marginal consumers like $o$ will substitute $B$ for $A$
$\uparrow q_A$ and overall $\uparrow Q_A$
Hence $\Gamma < 0$ implies $\frac{\partial Q_A}{\partial p_B}>0$ (goods are substitutes) (go back)

References

Anderson, Simon P., Simon Loertscher, and Yves Schneider. 2010. “The ABC of Complementary Products Mergers.” Economics Letters 106 (3): 212–15. doi:10.1016/j.econlet.2009.11.022.

Choi, Jay Pil. 2008. “Mergers with Bundling in Complementary Markets.” The Journal of Industrial Economics 56 (3). wiley: 553–77. doi:10.1111/j.1467-6451.2008.00352.x.

Flores-Fillol, Ricardo, and Rafael Moner-Colonques. 2011. “Endogenous Mergers of Complements with Mixed Bundling.” Review of Industrial Organization 39 (3): 231–51. doi:10.1007/s11151-011-9281-0.

Gentzkow, Matthew. 2007. “Valuing New Goods in a Model with Complementarity: Online Newspapers.” American Economic Review 97 (3): 713–44. doi:10.1257/aer.97.3.713.

Peitz, Martin. 2008. “Bundling May Blockade Entry.” International Journal of Industrial Organization 26 (1): 41–58. doi:10.1016/j.ijindorg.2006.09.005.

Ribeiro, Ricardo, and Joao Vareda. 2010. “Crowding Out or Complementarity in the Telecommunications Market.” Economics Letters 106 (3). Elsevier: 212–15.

Spector, David. 2007. “Bundling, Tying, and Collusion.” International Journal of Industrial Organization 25 (3): 575–81. doi:10.1016/j.ijindorg.2006.06.003.

Merger Effects with Product Complementarity: Evidence from Colombia’s Telecommunications

Intro

Product complementarity

Mergers with product complementarity

Recent examples: smartphones and headsets

Recent examples: phone and TV

Recent examples: Internet and cable TV

Recent examples: Comcel and Telmex

Recent examples: ETB and …

Motivation

Overview

Relation to literature

Data and industry

Colombia's telecom sector: the timeline

Colombia's telecom sector: the players

Data

Data: Form 5

Data Household survey

Summary statistics: firms

Summary statistics: plans

Summary statistics: correlations

Demand

Estimating substitution patterns

Empirical demand I

Empirical demand II

Empirical demand III

Empirical demand IV

Identification of Γ

Identification of Γ: intuition

Endogeneity and instruments

Instruments: examples

Supply side

How to do counterfactuals

Results

Results: product characteristics

Results: Susbtitution patterns

Results: Susbtitution patterns

Counterfactuals

Alternative scenarios

Merger ETB and Avantel

Merger ETB and Avantel (cont.)

De-merger of Claro

De-merger of Claro (cont.)

Final remarks

Thanks!

Summary statistics: households

Choice probabilities

Substitution patterns: ↑pB↑pB\uparrow p_B and Γ=0Γ=0\Gamma = 0

Substitution patterns: ↑pB↑pB\uparrow p_B and Γ>0Γ>0\Gamma > 0

Substitution patterns: ↑pB↑pB\uparrow p_B and Γ<0Γ<0\Gamma < 0

References

Merger Effects with Product Complementarity: Evidence from
Colombia’s Telecommunications

Substitution patterns: $\uparrow p_B$ and $\Gamma = 0$

Substitution patterns: $\uparrow p_B$ and $\Gamma > 0$

Substitution patterns: $\uparrow p_B$ and $\Gamma < 0$